[R-sig-ME] Problems fitting and interpreting glmmadmb and glmmTMB
Irene Rojo
ire.rojo at gmail.com
Fri Apr 13 11:08:34 CEST 2018
Hi,
I am trying to perform glmm analyses for fish density but I am getting
errors.. Also I am not an expert in statistics (at all) so I apologise if
my questions are too basic.
We sampled in 5 zones (ZN; fixed factor with 5 levels) and 3 protection
levels in each zone (PL; three levels). We selected 3, 6 and 9 sites (ST;
random effect) in each of the protection levels, respectively, and carried
out 3 underwater visual censuses in each site.
I am modeling counts of the most abundant species together as the response
variable, and including the area sampled as the offset term of the formula.
And I have so many zeros in my data.
I first tried the "glmer" function but it didn't work (model failed to
converge). Then I thought about the "zeroinfl" function but it doesn't deal
with random effects. It works well if I miss the random factor, but I don't
think that is right.
So I am trying to fit the models with the "glmmadmb" function as follows:
m0<- glmmadmb(nTRT10~ ZN*PL +
offset(log(areaTRT10))
+ ( 1 | ST),
data = den,
zeroInflation = TRUE,
family = "nbinom", link = "logit"
)
I am getting a huge error, either for the poisson or nbinom families, which
I don't understand:
Parameters were estimated, but standard errors were not: the most likely
problem is that the curvature at MLE was zero or negative
Error in glmmadmb(nTRT10 ~ ZN * PL + offset(log(areaTRT10)) + (1 | ST), :
The function maximizer failed (couldn't find parameter file)
Troubleshooting steps include (1) run with 'save.dir' set and inspect
output files; (2) change run parameters: see '?admbControl';(3) re-run with
debug=TRUE for more information on failure mode
In addition: Warning message:
running command 'C:\Windows\system32\cmd.exe /c glmmadmb -maxfn 500 -maxph
5 -noinit -shess' had status 1
However, with glmmTMB function it seems to be working well.
m0<- glmmTMB(nTRT10 ~ ZN*PL + ZN + PL +
offset(log(areaTRT10))
+ ( 1 | ST),
data = den,
zi=~ ZN + PL,
family = nbinom2,
dispformula = ~ PL
)
And I get this output:
Family: nbinom2 ( log )
Formula: nTRT10 ~ ZN * PL + ZN + PL + offset(log(areaTRT10)) + (1
| ST)
Zero inflation: ~ZN + PL
Dispersion: ~PL
Data: den
AIC BIC logLik deviance df.resid
1534.4 1626.1 -741.2 1482.4 226
Random effects:
Conditional model:
Groups Name Variance Std.Dev.
ST (Intercept) 0.09779 0.3127
Number of obs: 252, groups: ST, 84
Conditional model:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -6.0226 0.4126 -14.598 < 2e-16 ***
ZNCP 1.0949 0.4722 2.319 0.020405 *
ZNEF -0.1117 0.5170 -0.216 0.828938
ZNMN -1.8432 0.5088 -3.623 0.000291 ***
ZNTA -1.0188 0.5224 -1.950 0.051162 .
PLBZ -0.4027 0.5706 -0.706 0.480357
PLNT 1.1870 0.5262 2.256 0.024100 *
ZNCP:PLBZ 1.6528 0.6926 2.386 0.017015 *
ZNEF:PLBZ 1.8543 0.7333 2.529 0.011455 *
ZNMN:PLBZ 2.1438 0.7569 2.832 0.004624 **
ZNTA:PLBZ 1.4418 0.7361 1.959 0.050143 .
ZNCP:PLNT 0.4205 0.6577 0.639 0.522619
ZNEF:PLNT 1.2943 0.7438 1.740 0.081821 .
ZNMN:PLNT -0.1311 0.7308 -0.179 0.857588
ZNTA:PLNT 0.6470 0.7057 0.917 0.359197
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Zero-inflation model:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -0.6591 0.5512 -1.196 0.2318
ZNCP -19.2792 5421.1935 -0.004 0.9972
ZNEF -2.0983 0.9260 -2.266 0.0235 *
ZNMN -17.8418 4982.4109 -0.004 0.9971
ZNTA -20.4595 13048.1115 -0.002 0.9987
PLBZ 0.1975 0.7920 0.249 0.8031
PLNT -0.8759 1.0086 -0.868 0.3852
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Dispersion model:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -0.16608 0.21679 -0.766 0.444
PLBZ 0.09037 0.30420 0.297 0.766
PLNT 0.62906 0.32911 1.911 0.056 .
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
1. I don't understand the error in glmmadmb. Maybe the use of glmmTMB is
fine in this case so I don't need to use glmmadmb but still I am curious. Do
you think the use of glmmTMB is right here? I don't really understand the
differences among the two functions.
2. How can I perform model validation of glmmTMB? Is the same way as with
glmer?
3. How do I interpret the random effects conditional model?
Thank you so much,
Irene
[[alternative HTML version deleted]]
More information about the R-sig-mixed-models
mailing list